Convergence tests

[8813204602]

Dear Yambo developers and users
I want to do band structure calculation for ferromagnetic 2D NiCl2 using Yambo. I have a number of questions about the convergence tests need to do this:
1) I performed scf and nscf calculations using the vacuum size of 15 angestrom in the z direction using quantum espresso. Is it enough to consider the same value of 15 angestrom for the CUTBox variable or check the convergence of a variable such as the HF gap relative to this value?
2) It is correct to do the HF calculation and first to study the convergence tests of RandQpts and RandGvec parameters and then the convergence tests of FFTGvecs and EXXPLvcs parameters with respect to the HF gap, and then use the same converged values in the GW input file and just converge the remaining parameters such as NGsBlkxp, BndsRnxp,.... with respect to GW gap?
3) Is it right to study the convergence tests first with a small k-point sampling and then after obtaining the optional values of the parameters to check the convergence with respect to k-point? Wouldn't considering a small k-point have an effect on the optimized values?
I have another question:
For NiCl2 which is a two-dimensional ferromagnetic material with 3 atoms per unit cell, what do you think is the best value for nbnd in the nscf calculation for Yambo input? Is there a specific criterion for this choice?
Thanks a lot
Mitra Helmi,Ph.D student
Ferdowsi university of Mashhad
Mashhad,Iran

[Daniele Varsano]

Dear Mitra,

The forum is organised in subform to help the users to find relevant information in a easier way, please post your question in the correct subform, your questions are not related to compilation problem.

1) The box should be set slightly smaller than the cell size (i.e. 28 bohr would be ok). Note Yambo input units are in bohr.
2) Yes it is correct, note that FFTGvecs it is not a convergence parameter, if you not have memory problem you can leave at its default value, otherwise you can reduce it checking you don't loose accuracy.
3) Correct. K points are independent from the other convergence parameters and you can converge them as last variable once the other are converged
4) Empty bands: there is not a strict criterion, you can start considering 5/10 times the number of occupied bands and check if they are enough to converge the screening.

Best,

Daniele

PS: note you can fill your signature in your user profile, in that way you don't have to worry about signing up for every post

[8813204602]

Dear Daniele

Sorry for my post in the different subforum. I will certainly pay attention to this issue next time.
Thank you very much
Best regards,
Mitra

[8813204602]

Dear Daniele

In GW convergence tests, the two parameters BndsRnXp and NGsBlkXp must be converged simultaneously, as stated in the tutorial? Is it not possible to converge them separately?

Best regards,
Mitra Helmi

[Daniele Varsano]

Dear Mitra,
these two parameters are not independent, see e.g. Fig.1 of M. Stankovski et al. PHYSICAL REVIEW B 84, 241201(R) (2011).
There it is shown for bands in Sigma_c but the same happens for the bands in the response function.
Best,
Daniele

[8813204602]

Hello,

For k-point convergence, similar to other convergence tests, GW calculations can be performed for each new k-point mesh only for the k-point at which the gap is located, and then the convergence of the quasi-particle gap with respect to the k-points number is checked or for each new mesh, GW calculations must be performed for the all BZ?

Thanks a lot for your support

Best,
Mitra Helmi

[Daniele Varsano]

Dear Mitra,
I'm not sure I understood your question.
In general, you can check the convergence of the gap. Usually it is enough, but you can have different convergence for states having different character e.g. states with different localisation.

GW calculations must be performed for the all BZ?

This depends on the system and what do you want to calculate. In many case you can calculate the GW corrections for a set of states and see if a scissor/stretching approximation is valid. Note that calculations are performed on the irreducible BZ and not the full BZ as the full BZ is obtained by using the symmetry of the system.

Best,
Daniele

[8813204602]

Dear Daniele

Thanks a lot for quick response

That is, it is true that in the study of convergence with respect to k-point, instead of performing convergence tests with respect to the gap, convergence tests be performed with respect to VBM and CBM energies at a common point (e.g.gamma) for different nscf k-meshs. In other words, I can consider QPkrange as follows:
%QPkrange
1| 1| 17| 18|

Can it be said that because in different nscf k-meshs the k-point where the gap is located changes (for example for grid 6*6*1 the gap is at k-point 9 and for grid 9*9*1 the gap is at k-point 16), there is no common k-point for performing convergence test with respect to the gap.

Best,
Mitra

[Daniele Varsano]

Dear Mitra,

an easy way to overcome this is to make the convergence of the direct gap at Gamma point, this is common to all the non-shifted grid and in yambo it has always index 1.
Then, you can calculate the minimum gap using the converged parameters.

Best,
Daniele

[8813204602]

Dear Daniele,

Is it possible to use GTermKind alone without activating XTermKind, ie GTermKind =BG and XTermKind=none, or must the two be activated together?

Thank you for your support

Best,
Mitra